Soft-sensor Modeling Of Grinding Process Based On Neural Network And Adaptive Revision

In the grinding and classifying process of iron ore,some key variables need to be tracked and monitored.Soft-sensor modeling has become the mainstream method for monitoring key variables in industrial processes.Aiming at the complex industrial environment in the grinding and classifying process,the continuous change of industrial state and the strong nonlinear relationship,this paper proposes three soft-sensor models to monitor the key variables in grinding and classification process.(1)In the case of a relatively stable industrial environment,the extreme learning machine(ELM)soft-sensor model with different activation functions on grinding process optimized by improved black hole(BH)algorithm was proposed.The kernel principal component analysis(KPCA)method is first used to reduce the dimensionality of high-dimensional data;In order to investigate the influence of different activation functions on the prediction accuracy of the ELM model,seven continuous function(Arctan,Hardim,Morlet,Re Lu,Sigma,Sin and Tanh)are used as the activation function of the ELM neural network to establish soft-sensor models respectively.For the shortcomings that ELM model weights and biases values are arbitrarily given so as to result in the low prediction accuracy and low reliability,an improved BH algorithm based on the golden sine operator and the Levy flight operator(GSLBH)was used to optimize the parameters of the ELM neural network.Simulation results show that the model has better generalization and prediction accuracy,and can meet the real-time control requirements of the grinding process.(2)Least squares support vector machine(LSSVM),as a soft-sensor model with strong generalization ability,can be used to predict key production indicators in complex grinding processes.The traditional crossvalidation method cannot obtain the ideal structure parameters of LSSVM.In order to improve the prediction accuracy of LSSVM,a golden sine Harris Hawk optimization(GSHHO)algorithm was proposed to optimize the structure parameters of LSSVM models with linear kernel,sigmoid kernel,polynomial kernel,and radial basis kernel,and the influences of GSHHO algorithm on the prediction accuracy under these LSSVM models were studied.In order to deal with the problem that the prediction accuracy of the model decreases due to changes of industrial status,this paper adopts moving window(MW)strategy to adaptively revise the LSSVM(MW-LSSVM),which greatly improves the prediction accuracy of the LSSVM.The prediction accuracy of the regularized extreme learning machine with MW strategy(MW-RELM)is higher than that of MW-LSSVM at some moments.Based on the training errors of LSSVM and RELM within the window,this paper proposes an adaptive hybrid soft-sensing model that switches between LSSVM and RELM(MW-RELMLSSVM).Compared with the previous MW-LSSVM,MW-neural network trained with extended Kalman filter(MW-KNN),and MW-RELM,the prediction accuracy of the hybrid model is further improved.Simulation results show that the proposed hybrid adaptive soft-sensor model has good generalization ability and prediction accuracy.(3)Long and short-term memory model(LSTM)has been used in online soft sensor modeling of industrial processes in recent years.However,the appearance of abnormal points and random noise in the samples will greatly impair the performance of the global LSTM model,and random changes in the industrial state will cause the global LSTM model to easily discard critical data information.In order to solve the above two problems encountered when using global LSTM modeling,this paper proposes a variational autoencoder bidirectional LSTM soft sensor modeling method based on batch training(Bt-VAEBi LSTM).First,the training sample set is divided into multiple batches according to the time series,in order to reduce the influence of abnormal points and random noise on the prediction accuracy of Bi LSTM,the variational autoencoder(VAE)is then used to extract the data features of the training samples in each batch,and reconstruct the training samples according to these data features,the reconstructed sample can more clearly reflect the current industrial status.In order to solve the problem of the global LSTM model discarding critical data information during training,this paper proposes a batch training method,the reconstructed samples are trained in batches according to the time series.After the training of each batch samples is completed,the structural parameters of the previous local Bi LSTM model are shared with the next local Bi LSTM model as the initial parameters to ensure that important industrial status information will not be lost.At the same time,in order to prevent the Bt-VAEBi LSTM model from overfitting,this paper introduces the L2 regularization term in the loss function.The effectiveness of the proposed method is verified by simulation experiments on the grinding and classifying process.